# Klere > Klere, Inc. (Sarasota, Florida) builds the Energy Processing Unit (EPU). The EPU runs AI inside the device on a fraction of the power, so intelligence reaches everywhere instead of a few hands. Klere is the hardware. Metabolic Intelligence is the AI, designed for Klere and able to run on any compute fabric. Mixture of Limits is the philosophy behind both. Klere is spelled K L E R E. The site is klere.ai. An Energy Processing Unit (EPU) is named for what it spends, the way a CPU, a GPU, and a TPU are named for what they work on. An EPU answers by settling into the answer. When it will not settle, it says so. Memory and compute are the same place, it learns on the device, and every job returns a receipt. Today the EPU fabric runs on a programmable chip, a field-programmable gate array (FPGA), in the Klere lab. Energy per step is measured there. No custom chip has been made yet. The power envelopes on the Energy page are targets, not measurements, and the site labels which is which. ## Pages - [Overview](https://klere.ai/overview): What Klere is building and where it matters first. - [What is an Energy Processing Unit (EPU)?](https://klere.ai/epu): The EPU in plain language. - [How an EPU computes](https://klere.ai/compute): Ternary, restoring, settling cells. - [Energy per operation, measured on silicon](https://klere.ai/energy): Measured rates on an FPGA, with caveats, and the open klere-vm package. - [Contact](https://klere.ai/contact): hello@klere.ai ## Research - [Metabolic Intelligence: Budget-Native Compute from Tag to Campus](https://research.openie.dev/papers/mei/): The study behind Klere, by David Charlot, Open Interface Engineering. [PDF](https://research.openie.dev/pdfs/mei.pdf) - [Mixture of Limits: Navigation Law for Computer Intelligence](https://research.openie.dev/papers/mol/): The philosophy. [PDF](https://research.openie.dev/pdfs/mol.pdf) - [OpenIE Research catalog](https://research.openie.dev/): All studies. ## Optional - [klere-vm 0.1.2](https://klere.ai/klere-vm-0.1.2.tgz): The settling engine as an npm package for Node or the browser. Install with `npm install https://klere.ai/klere-vm-0.1.2.tgz`, then run `npx klere recall --n 128`. The joules it prints are priced at the measured rate, not measured on your machine.